Microsoft has deployed its artificial intelligence (AI)-powered MDASH security scanner to Azure Government, giving select U.S. government customers and authorized partners preview access to technology designed to identify and validate exploitable vulnerabilities in software.
The company said in a Sept. 8 blog post that its multi-model agentic scanning system – codenamed MDASH – is designed to uncover complex software vulnerabilities that traditional pattern-based security scanning tools may miss. For government agencies, Microsoft said the technology could help security teams identify weaknesses in mission-critical software and its broader supply chain before cyberattackers exploit them.
MDASH works as an agentic code scanner within Microsoft Defender, analyzing source code to determine whether suspected vulnerabilities are reachable and exploitable.
Unlike traditional scanning tools that generally search for known patterns associated with software flaws, MDASH uses more than 100 specialized AI agents and multiple AI models to analyze the same codebase, according to Microsoft.
“We can agentically do things now to rationalize data that we weren’t able to, and it doesn’t put as much burden on the end user,” Steve Faehl, Microsoft’s chief technology officer for US Public Sector, said of the new tool while speaking at the Billington CyberSecurity Summit on Sept. 8.
“If we can lift that burden away, end users can be more productive and also more secure at the same time,” Faehl added.
Each agent specializes in identifying a particular category of software weakness, according to the company. The system then passes potential vulnerabilities to another group of agents that evaluates the evidence for and against whether each flaw is reachable and poses a genuine security risk.
Microsoft said the system also consolidates duplicate findings and, where possible, demonstrates that a vulnerability can be exploited rather than simply flagging a potential weakness. The process is intended to provide security teams with a more refined and prioritized set of vulnerabilities to investigate and remediate.
At the center of MDASH is what Microsoft calls a harness, which directs and coordinates different AI models based on the task being performed. Microsoft said the architecture allows the system to incorporate new AI models without requiring agencies to rebuild the broader scanning environment.
Using multiple models also provides independent assessments of potential vulnerabilities. Microsoft said confidence in a finding increases when multiple models reach the same conclusion or when one model identifies a potential flaw that another model cannot dismiss.
The architecture is also intended to lower scanning costs as AI models improve. Microsoft said the newest addition to its MAI model family is expected to reduce the cost of an individual scan by roughly half.
Microsoft said agencies have been prioritizing software for review based on mission importance because the amount of code requiring in-depth analysis exceeds current scanning capacity. Reducing the cost of scans, the company said, could move agencies closer to reviewing 100% of their code without requiring additional budget cycles.